IUBio

Memory (and SOMs)

mat mats_trash at hotmail.com
Mon Feb 18 15:36:18 EST 2002


> My books have just arrived and I have a better understanding of SOMs
> now =)  From what I read, SOMs extract the "principle components"
> of a set of data that is presented to the network repeatedly over a
> period of time (the training). 

>From the definition of how an SOM works the nodes in a network will be
defined by the common elements in the training information which have
been categorised to each particular node. The modal information in the
training datasets (i.e. those pieces of information that occur most
often in the datasets) will have the overriding effect on defining the
node (after all the node is defined by the information classified to
it and nothing else)Subsequent test data that has part or all of this
modal information will thus be classified to the node.  (This is my
understanding of how SOMs work anyway)

>  The essence of this process is one of
> *abstraction*. Upon presenting 10 cows, an SOM will respond with
> the quintessential "cow", but is that what memory does? Memory is
> more like, presenting a cow evokes the memory of a farm scene
> in the past, which is more like an associative network and is
> excitatory in nature (versus inhibitory as in winner-takes-all).


Yes but what if I asked you to describe a cow?  You would describe the
defining elements no?  Similarly, what separates the nodes in an SOM
is the difference in information which defines them thus SOMs store
the salient and defining charachteristics of datasets presented to
them.

Incidentally, it is the SOM architecture (and modifications of it)
that is currently being used to speed up throughput in applications
such as reporting on biopsy specimens from patients with suspected
cancer.  After long training, SOMs come to be able to classify
specimens given as they can extract the salient and defining
charachteristics.  Of course, in terms of human memory I would suppose
that the neuronal functional architecture cannot be simulated by
'simple' neural net models, especially as brain function is likely to
be stochastic and chaotic.




More information about the Neur-sci mailing list

Send comments to us at biosci-help [At] net.bio.net